Recovery scores explained, and why your devices disagree

In one night at a Central Queensland University sleep lab, Miller and colleagues (2022) fitted 53 adults with six consumer wearables plus an electrocardiogram and compared every device's heart rate variability with the ECG. WHOOP 3.0 read 4.5 ms low on average. Apple Watch Series 6 read 9.6 ms low. A Garmin Forerunner 245, which took its reading from a short sample near the start of the night, read 22.4 ms low. Those gaps exist before any scoring formula touches the data. Stack a different baseline, a different set of extra inputs and a different scale on top, and the same person can wake up to a 7 out of 10, a 64 percent and a 38 on three screens.

None of those numbers has to be wrong. Each answers a slightly different question. This article covers what WHOOP, Oura, Garmin and Apple say about their scores, how Titan's Recovery score works down to the weights with three mornings calculated in full, what the research supports, and what to change on a Low, Moderate or High day.

01What every score starts from

Every major recovery or readiness score draws on heart rate variability, and most pair it with overnight resting heart rate. Each company compares those signals with your own history instead of a population norm. The differences between products come later, in which extra inputs they add, how long a history they compare against, and how they turn the comparison into a number.

Heart rate variability

HRV is the beat-to-beat variation in the interval between heartbeats. At rest most of the fast variation comes from the vagus nerve, the parasympathetic brake on the heart, so HRV above your usual level generally means a rested system. RMSSD, the root mean square of successive differences between beats, tracks vagal activity closely. SDNN, the standard deviation of all intervals in a recording, reflects a broader mix of influences. WHOOP reports RMSSD. Apple Watch stores HRV in Apple Health as SDNN, as the WHOOP support and Recovery settings articles document.

The raw millisecond value says little on its own. Age, genetics, fitness, breathing rate and the measurement window all shift it, so 45 ms can be a great morning for one person and a poor one for another. Plews and colleagues (2013) argued that only longitudinal monitoring reveals each athlete's individual HRV pattern, which is why every serious score compares you with yourself.

Resting heart rate

Resting heart rate usually moves opposite to HRV. Heat, dehydration, alcohol, a late meal, illness and accumulated training fatigue tend to raise overnight heart rate and lower HRV. HRV reacts faster and noisier, and resting heart rate is steadier night to night, so when both move against you on the same morning the signal is harder to dismiss. Plews and colleagues (2013, IJSPP) found that a single day's resting heart rate correlated weakly to moderately with training gains in 10 runners, and a one-week average correlated strongly. The same was true, more sharply, for HRV.

02How WHOOP, Oura, Garmin and Apple build their numbers

Each summary below uses only what the manufacturer publishes about its own score. None of the four publishes its weights. Some publish band cutoffs and some do not.

WHOOP Recovery

WHOOP's developer documentation describes Recovery as "a daily measure of how prepared your body is to perform," calculated when you wake as a percentage from 0 to 100. The higher the score, the more ready WHOOP considers you to take on Strain that day. The documentation lists the objective measurements that factored into the score as resting heart rate, HRV as RMSSD and, on WHOOP 4.0, blood oxygen and skin temperature. Its data model includes a user_calibrating flag for the period while WHOOP is still learning your baseline, and a score state of "UNSCORABLE" when there is not enough data.

The documentation publishes no weights or band cutoffs, and I could not reach a WHOOP page that states them, so this article quotes none. WHOOP also does not write HRV to Apple Health, so a WHOOP band alone cannot produce Titan's Recovery score, as the WHOOP support article explains.

Oura Readiness

Oura's Readiness Score runs from 0 to 100 with published bands. At 85 or higher Oura calls it "Optimal, you're ready for action!" From 70 to 84 it is "Good, you've recovered well enough." Under 70 it is "Pay attention, you're not fully recovered." Oura groups its contributors into three pillars and lists them by name.

• Sleep and Sleep Balance, covering last night and the past two weeks
• Previous Day Activity and Activity Balance, covering yesterday and recent load
• Resting Heart Rate, HRV Balance, Body Temperature and Recovery Index, which Oura groups as body stress

Recovery Index asks how many hours of sleep you got after your heart rate reached its overnight low. Oura's page states that the contributors sum to the score, stresses the long-horizon contributors, and says that "100s are designed to be rare rather than regular." Oura's Readiness therefore folds sleep and activity directly into the number. A short night or an unusually active yesterday can lower Readiness even when HRV and resting heart rate look normal.

Garmin Training Readiness and Body Battery

Garmin ships two related numbers. Training Readiness is the morning-facing one. Garmin names last night's sleep and the recovery time left from recent activities as the primary drivers, followed by acute training load, HRV status, sleep history before last night and stress history from the past three waking days. Its HRV status reads as balanced when your 7-day average HRV sits within your personal baseline range. Recent activities count fully toward acute load when recorded, and their effect expires over the next 10 days. Garmin classifies readiness in five levels, "from poor to prime with low, moderate, and high in between," and updates it through the day as recovery time runs down or a new workout lands. Garmin's page does not give numeric cutoffs for those levels, and it says Training Readiness is not designed to predict race performance.

Body Battery answers a different question. Garmin describes it as a running estimate of energy reserves built from heart rate, HRV and movement, fullest when you wake, drained by activity and stress, and recharged by rest and sleep. The two can disagree on the same morning because Training Readiness looks back across 10 days of load and several nights of sleep, while Body Battery tracks the current day.

Apple Readiness and Vitals

Apple added a Readiness score with Apple Watch Series 12 and Ultra 4, announced on September 9, 2026. Apple says it analyzes recent activity, training load, vitals and sleep score, and returns a single score from 0 to 10 with one of four recommendations, Recover, Pace Yourself, Ready or Go For It. Readiness updates through the day, for example after an intense workout, and shows which factors are driving the score. Apple has not published the score ranges behind each recommendation.

The Vitals app remains the per-metric view. Apple's announcement says overnight vitals now include a recovery HRV measurement analyzed against your personal baseline, labeled in words such as "Typical" and "Favorable," and adds a daytime view. The Apple Watch HRV that Titan reads from Apple Health is SDNN.

The five scores side by side

Oura, Garmin and Apple put sleep and training load inside the number. WHOOP and Titan stay close to the cardiac signals and leave training load to separate scores. That design choice is the first place to look when two devices disagree.

03How Titan's Recovery score works

Everything in this section comes from the Recovery score and baselines help articles, which document the model in full.

The two inputs

Recovery uses one HRV value and one resting heart rate value per day. By default the HRV value comes from the first of three sources that has data. First is a Mindfulness session recorded before noon. Next is the average HRV during your night's sleep. Last is the average of all HRV from the past 24 hours. Turning off Prioritize Sleep or Mindfulness HRV under Settings > Recovery & Sleep > Recovery Preferences makes Titan use the all-day average instead. Resting heart rate is today's value from Apple Health, falling back to your average over the last 3 days, then the last 7.

The baseline, in median absolute deviations

Your baseline is the median of your daily HRV and the median of your daily resting heart rate across your Recovery baseline window. You choose 7, 30 or 60 days during onboarding, and 60 is the default. The median is the middle value, so one strange night barely moves it.

Titan measures today's distance from that median in median absolute deviations, or MADs. The MAD is the median of how far each day in the window sits from the median, a measure of your normal day-to-day spread that one outlier cannot drag around. Scaling by it means a 5 ms rise counts for more when your HRV rarely moves. For an athlete whose HRV MAD is 4 ms, a 5 ms rise is 1.25 MADs. For one whose MAD is 12 ms, the same rise is about 0.4 MADs, so it moves the first athlete's baseline term about three times as far.

Five weighted signals

Titan scores five signals and adds them with fixed weights.

Positive weights reward a higher value and negative weights penalize one. HRV above baseline raises the sum, resting heart rate above baseline lowers it, and a more erratic HRV history lowers it slightly.

The logistic curve and the percentile step

Titan passes the weighted sum through a logistic curve, the standard S-shaped function 1 / (1 + e^−x), which returns a value between 0 and 1. A sum of 0 returns 0.5. Large positive sums approach 1 and large negative sums approach 0.

Titan then places that value against your last 60 Recovery calculations. Your 10th percentile maps to 0 and your 90th percentile maps to 100, with values in between placed along the line joining those two anchors. Anything beyond an anchor reads 0 or 100. A score of 100 therefore means today sits in your own top tenth of recent mornings, and a 0 means your bottom tenth. With a stable history, roughly one morning in ten will read 100 and one in ten will read 0 by construction. A 100 means one of your best mornings in two months of history. It says nothing about your HRV in absolute terms.

Three mornings, worked through

Take an Apple Watch user on the default 60-day window. Her baseline HRV median is 62 ms with a MAD of 8 ms. Her baseline resting heart rate median is 52 bpm with a MAD of 2 bpm. The coefficient of variation of her daily HRV is 0.18. Across her last 60 calculations, the 10th percentile logistic value was 0.05 and the 90th was 0.97.

Check Morning B by hand. Its logistic value of 0.663 sits 0.613 above the 0.05 anchor, across a 0.92 span to the 0.97 anchor. That is 67 percent of the way, so the score reads 67.

What the arithmetic shows

The weekly HRV term carries most of the swing. It is measured in raw milliseconds, while the baseline term is measured in MADs. Each millisecond is worth 0.5 through the weekly term and 1 divided by your MAD through the baseline term, so for anyone whose HRV MAD is above 2 ms, which is nearly everyone, the weekly comparison counts for more. On Morning A, an 8 ms drop against the week contributed −4.00, against −1.50 from sitting 1.5 MADs below the 60-day median. Resting heart rate works the other way. Its MAD is usually a beat or two, so the baseline term at 0.7 per MAD outweighs the weekly term at 0.3 per beat. Titan's Recovery therefore reacts most to how today's HRV compares with the past week, and to how today's resting heart rate compares with your longer normal. This is also why switching Titan to RMSSD, which sits on a different millisecond scale from Apple's SDNN, calls for rebuilding history with Clear Recovery Cache, as the Recovery settings article instructs.

The logistic curve saturates on big days. Sums beyond about ±5 land within a percent of 0 or 1, so clear good and bad mornings pin to the ends of the scale, and the percentile step does the separating in the middle.

The variability term is small. At a coefficient of variation of 0.18, it moves the sum by less than a tenth, and it changes slowly because it is computed across many days. Of the five weights, this is the one I would hold most loosely against the literature, which the evidence section covers.

The first weeks and the fallback

The full model runs only when the baseline window holds at least 7 days of both HRV and resting heart rate, with readings on at least 60 percent of its days. A 7-day window needs all 7 days, a 30-day window needs 18 and a 60-day window needs 36.

Until then, Titan shows a rough fallback built from raw values. The HRV part is your HRV in ms, capped at 100. The resting heart rate part starts at 100 at 40 bpm and drops 2 points for each beat above. The score is the average of the two. HRV of 50 ms and resting heart rate of 55 bpm give 50 and 70, so the fallback reads 60. The fallback has no idea what is normal for you, so someone with a naturally low HRV will read low every day until the full model takes over, and the number usually shifts at that point.

04What the evidence supports

Three questions matter. Do HRV trends track training adaptation, does adjusting training to HRV beat a fixed plan, and do consumer sensors measure the signal well enough to begin with?

HRV trends track adaptation, with caveats

Plews and colleagues (2013, Sports Medicine) reviewed HRV monitoring in elite endurance athletes and found the picture messier than a simple rule. Both increases and decreases in HRV have accompanied poor adaptation in elites, and fitness gains have sometimes arrived with an unexpected fall in HRV. Their fixes were methodological. Average across days, use indices that suit highly trained athletes, and learn each athlete's pattern over time.

The averaging point has direct evidence. In 10 runners over a 9-week block, a single day's log-transformed RMSSD had a trivial correlation with the change in maximal aerobic speed (r = −0.06), while the one-week average correlated strongly (r = 0.72) (Plews et al., 2013, IJSPP). A follow-up in triathletes found that at least 3 valid readings per week were needed to match the full weekly average (Plews et al., 2014). Titan's use of a multi-week median and a 7-day median follows the same logic.

Buchheit (2014) argued that most contradictory findings in this field come from methodological inconsistency and misinterpretation, and that any change should be judged against the measurement's error and the smallest worthwhile change, in the context of the current training phase. He also recommended pairing heart rate measures with training logs, questionnaires and simple performance tests, since no single marker captures every kind of fatigue.

Bellenger and colleagues (2016) pooled 24 studies in endurance athletes. Training that improved performance produced a small rise in resting RMSSD (standardized mean difference 0.58). Training that left athletes overreached and slower also produced a small rise (0.26). Their conclusion was that resting HRV is largely unaffected by overreaching. A single morning's HRV cannot tell you whether a hard block is working or digging a hole. The pattern over days, alongside how you feel and perform, carries more information.

On variability, the evidence is thin and points in more than one direction. In a two-athlete case study over 77 days, Plews and colleagues (2012) watched the 7-day average of log RMSSD fall steadily in the triathlete who became non-functionally overreached, while the coefficient of variation of that average also fell. The control athlete stayed stable on both. Titan's model penalizes higher variability. Two athletes cannot settle the direction, and the small size of Titan's variability term limits the stakes.

HRV-guided training in controlled trials

Kiviniemi and colleagues (2007) randomized 26 moderately fit men to a predefined plan, an HRV-guided plan or a control group for 4 weeks. The HRV group trained hard when morning HRV held steady or rose, and trained easy or rested when HRV fell below a reference derived from the previous 10 days or declined for 2 days in a row. VO2 peak rose from 56 to 60 ml/kg/min in the HRV group and did not change significantly in the predefined group, 54 to 55. Maximal running speed improved more in the HRV group. The between-group difference in VO2 peak change was not significant.

Vesterinen and colleagues (2016) ran 40 recreational runners through 8 weeks of intensive training. The HRV-guided group scheduled moderate and hard sessions only when morning HRV sat within an individual smallest worthwhile change, and ended up doing fewer of them, 13.2 against 17.7. Their 3,000 m running performance improved by 2.1 percent, a significant change, while the predefined group's 1.1 percent improvement was not. VO2 max improved in both groups.

Javaloyes and colleagues (2019) did the same with 17 well-trained cyclists over 8 weeks. The HRV-guided group improved peak power output by 5.1 percent, power at the second ventilatory threshold by 13.9 percent and a 40-minute time trial by 7.3 percent. The traditional group did not improve significantly, although the direct between-group comparison was not significant either.

Düking and colleagues (2021) pooled 8 studies with 198 participants. HRV-guided training had a significant positive effect on submaximal physiological markers (Hedges' g = 0.296) and small, non-significant effects on performance (g = 0.079) and VO2 peak (g = 0.171). HRV-guided plans usually contained fewer moderate and hard sessions, and they produced fewer non-responders.

Read together, these trials support a modest claim. Using morning HRV to decide when to go hard produces results at least as good as a fixed plan, often with fewer hard sessions and fewer athletes who fail to improve. The samples are small, dozens of participants per trial, and none of them tested a consumer composite score. They tested the practice of adjusting intensity to a personal HRV trend, a form of auto-regulation, which is what a recovery score is for.

How accurate the sensors are

Miller and colleagues (2022) is the clearest head-to-head. Against ECG, intraclass correlations for HRV were 0.99 for WHOOP 3.0, 0.67 for Apple Watch Series 6, 0.65 for Polar Vantage V, 0.63 for Oura Gen 2 and 0.24 for Garmin Forerunner 245. For heart rate the figures were 0.99, 0.96, 0.93, 0.85 and 0.41. Two caveats apply. The research group receives funding and equipment from WHOOP, and WHOOP supplied raw beat-to-beat data where the other devices were read from their apps. The Garmin HRV came from a 3-minute window near sleep onset, so its low agreement partly reflects when it measured.

Dial and colleagues (2025), with no declared conflicts, tracked 13 adults over 536 nights. Lin's concordance with ECG for overnight HRV was 0.99 for Oura Gen 4, 0.97 for Oura Gen 3, 0.94 for WHOOP 4.0, 0.87 for Garmin Fenix 6 and 0.82 for Polar Grit X Pro. Resting heart rate concordance was 0.98 and 0.97 for the two Oura rings, 0.91 for WHOOP and 0.86 for Polar. The two studies used different hardware generations and statistics, so compare devices within a study, not across them. Both show that two wearables on the same body report different HRV on the same night before any scoring model runs.

05Why your number disagrees with another device or another person

Another person

A friend's Recovery of 80 and your 45 say nothing about who is fitter or healthier. Each score measures distance from that person's own history. Titan's percentile step makes this explicit, since your 100 is your top tenth and theirs is theirs. The raw HRV values behind the scores are not comparable either, for the reasons in the HRV section above.

Another device

When two devices disagree on the same morning, the causes stack in a predictable order.

1. Measurement. The devices may record different HRV statistics, SDNN on Apple Watch and RMSSD on WHOOP, over different windows, a whole night, a sleep phase, or a short morning reading. Sensor agreement with ECG also differs, as Miller and Dial showed.
2. Inputs. Oura, Garmin and Apple fold sleep and training load into readiness. Titan and WHOOP stay close to the cardiac signals. A poor night of sleep with normal HRV will lower Oura's or Garmin's number more than Titan's.
3. Baseline length. Garmin's HRV status compares a 7-day average with a personal range. Oura's balance contributors look back two weeks or more. Titan compares with a 7, 30 or 60-day median. A longer window treats a rough training week as a real drop, and a short one absorbs it.
4. Scaling. Titan ranks today against your last 60 scores. Oura uses a 0 to 100 scale with fixed bands. Apple uses 0 to 10 and four labels. Garmin uses five named levels.
5. Timing. Apple and Garmin update readiness through the day. Titan's Recovery can change after noon if it started from a Mindfulness reading, unless you freeze it with Enable Titan to Save Resting Heart Rates.

Titan's baselines article shows how much the window alone matters.

Another screen inside Titan

Titan's own numbers use different baselines too. Recovery and the Stress headline on Today use your Recovery window. Battery divides recent HRV by a fixed 60-day median whatever window you chose, and the Apple Watch app always scores against the 60-day default. If you chose a 7-day Recovery window, Recovery and Battery can point in different directions on the same morning.

Yesterday's score changed

Titan recalculates up to 60 days of Recovery each time Today refreshes, against the baseline window that ends today, so older scores shift as new days enter the window. The why past scores changed article lists every cause.

06What to do on a low, moderate or high day

A recovery score is an input to a training decision. The research above supports adjusting intensity to a personal trend, so the useful questions are what today's band says and whether the pattern has held for more than one morning.

Low day

Titan's coaching text for a Low score, 0 to 39, reads "Your recovery is low today. Consider reducing intensity, prioritizing sleep, and focusing on easy aerobic work." Keep the session and lower the ceiling. Swap intervals for easy aerobic volume, cut the top sets from a strength session, or make it a rest day if the week has been heavy. Several Low days inside a hard block are a reason to bring a deload week forward.

One Low morning after a hard session is expected. A run of them deserves more weight. Titan does not use a fixed millisecond or percentage HRV threshold. For a rule of thumb that works with any wearable, use this one. A morning HRV more than one median absolute deviation below your baseline median for two or more consecutive days, with resting heart rate above baseline, is a signal to pull intensity. The Recovery score and baselines articles show where to find your median and how your window behaves. Kiviniemi's protocol also acted on a two-day decline, and the averaging studies above show why a single reading is too noisy to act on.

Moderate day

The Moderate band, 40 to 74, reads "Your recovery is moderate. You can train, but keep an eye on intensity and make sure you support recovery with sleep and nutrition." Moderate is where most ordinary training days land. Run the planned session and watch how the hard efforts feel against their usual heart rate or pace.

Moderate days are where the weekly view helps most. Recovery-Based Training in Trends plots a 7-day rolling average of Exertion times 10 minus Recovery. Titan treats −10 to +10 as the Optimal Training Range. Above +10 is Overreaching, where "exertion is outpacing recovery," and the guidance is to consider tapering intensity or adding a lighter day. Below −10 is Restorative, and the added line depends on your training goal. A string of Moderate mornings is fine while that line stays in range. A string of Moderate mornings with the line climbing past +10 is the pattern to act on.

High day

The High band, 75 to 100, reads "Your recovery is high. This is a good day for quality training if it aligns with your plan." Put the key interval session or heavy lift here when the plan allows. Keep in mind what the percentile step means. A High score says today ranks near the top of your last 60 mornings, measured by HRV and resting heart rate alone. It does not see a sore hamstring, a sore throat, a long-haul flight or a poorly healed injury. Any of those overrides the number.

07When the number is missing or wrong

Some bad scores are data problems. The Recovery missing or looks wrong article covers each cause in order.

• The baseline is still filling. Until your window meets the coverage rule, 36 days of data for the default 60-day window, Titan shows the fallback score.
• Today has no HRV or no resting heart rate. Recovery stays blank. Wear the watch to sleep and check Health permissions for both types. HRV not showing covers the HRV side.
• The score changed during the day. A morning score built from a Mindfulness session can change after noon, when Titan moves to sleep HRV.
• You switched to RMSSD. Titan then needs the Heartbeat Series permission, and without it HRV and Recovery can go blank. Clear the Recovery cache after any change of HRV method.
• Past scores look stale. Clear Recovery Cache under Settings > Health & Connected Apps > Resync & Troubleshooting recalculates Recovery, Sleep, Stress and Battery from Apple Health.

08The Recovery detail screen on Today

Tap the Recovery card on Today to see your own inputs, baselines and history. The top shows your score in a ring, the band label and the time your sleep ended. The "What shapes your score" section shows today's HRV and resting heart rate beside your baselines, with the difference in ms or bpm. Tap either row for the baseline window length, the percent change from baseline and a daily chart. Titan notes that these rows show which way each input moved and are not exact contributions to the score. The "Recovery history" section has a 7, 30 and 60 day picker, a daily bar chart, and a comparison of your last 7 days with the 7 before once each week has 4 scored days. The Recovery score article documents the full screen.

09References

• Bellenger CR, et al. (2016). Monitoring athletic training status through autonomic heart rate regulation: a systematic review and meta-analysis. Sports Medicine, 46(10), 1461-1486. doi.org/10.1007/s40279-016-0484-2
• Buchheit M. (2014). Monitoring training status with HR measures: do all roads lead to Rome? Frontiers in Physiology, 5, 73. doi.org/10.3389/fphys.2014.00073
• Dial MB, et al. (2025). Validation of nocturnal resting heart rate and heart rate variability in consumer wearables. Physiological Reports, 13(16), e70527. doi.org/10.14814/phy2.70527
• Düking P, et al. (2021). Monitoring and adapting endurance training on the basis of heart rate variability monitored by wearable technologies: a systematic review with meta-analysis. Journal of Science and Medicine in Sport, 24(11), 1180-1192. doi.org/10.1016/j.jsams.2021.04.012
• Javaloyes A, et al. (2019). Training prescription guided by heart-rate variability in cycling. International Journal of Sports Physiology and Performance, 14(1), 23-32. doi.org/10.1123/ijspp.2018-0122
• Kiviniemi AM, et al. (2007). Endurance training guided individually by daily heart rate variability measurements. European Journal of Applied Physiology, 101(6), 743-751. doi.org/10.1007/s00421-007-0552-2
• Miller DJ, et al. (2022). A validation of six wearable devices for estimating sleep, heart rate and heart rate variability in healthy adults. Sensors, 22(16), 6317. doi.org/10.3390/s22166317
• Plews DJ, et al. (2012). Heart rate variability in elite triathletes, is variation in variability the key to effective training? A case comparison. European Journal of Applied Physiology, 112(11), 3729-3741. doi.org/10.1007/s00421-012-2354-4
• Plews DJ, et al. (2013). Training adaptation and heart rate variability in elite endurance athletes: opening the door to effective monitoring. Sports Medicine, 43(9), 773-781. doi.org/10.1007/s40279-013-0071-8
• Plews DJ, et al. (2013). Evaluating training adaptation with heart-rate measures: a methodological comparison. International Journal of Sports Physiology and Performance, 8(6), 688-691. doi.org/10.1123/ijspp.8.6.688
• Plews DJ, et al. (2014). Monitoring training with heart rate-variability: how much compliance is needed for valid assessment? International Journal of Sports Physiology and Performance, 9(5), 783-790. doi.org/10.1123/ijspp.2013-0455
• Vesterinen V, et al. (2016). Individual endurance training prescription with heart rate variability. Medicine & Science in Sports & Exercise, 48(7), 1347-1354. doi.org/10.1249/MSS.0000000000000910
• Apple (2026). Introducing Apple Watch Series 12, with the all-new Health Sensing System. Apple Newsroom. apple.com/newsroom
• Garmin (2024). Training Readiness. Garmin Technology. garmin.com
• Garmin (n.d.). Body Battery. Garmin Technology. garmin.com
• Oura (2026). Your Oura Readiness Score. Oura blog. ouraring.com
• WHOOP (n.d.). Recovery. WHOOP for Developers. developer.whoop.com